Smart Venue Map Skill

Generate a fully offline, single-file HTML indoor map with:

• IMDF floor plan rendered via Leaflet.js (bundled inline)
• Per-floor navigation; units colored by category
• Manual sensor/camera placement (persisted in localStorage)
• Zone group management with aggregated analytics popups
• Import/export of placement data as JSON

Inputs the user may provide

If VA data is missing, the map still works — sensors show "No data linked".

Step 1 — Install dependencies

python3 -m pip install --user openpyxl  # only if VA data is .xlsx
npm install leaflet                    # provides leaflet.js + leaflet.css

Verify Leaflet was installed:

ls node_modules/leaflet/dist/leaflet.js && echo "OK"

Step 2 — Process IMDF + VA data

Use the bundled script: scripts/process_imdf.py

# With Excel VA data
python3 {baseDir}/scripts/process_imdf.py \
  --imdf-zip "path/to/IMDF.zip" \
  --va-xlsx  "path/to/va_data.xlsx" \
  --out-dir  ./outputs

# Without VA data (map only)
python3 {baseDir}/scripts/process_imdf.py \
  --imdf-zip "path/to/IMDF.zip" \
  --out-dir  ./outputs

Outputs:

• outputs/imdf_b64.txt — gzip+base64 IMDF for embedding
• outputs/dash_data.json — processed sensor analytics data
• outputs/imdf_compact.json — intermediate (for inspection)

Column auto-detection: The script recognises common column names (camera_id, date, hour, gender, age_group, eyewear). If the user's Excel has different column names, read the file first and pass the correct field names or patch the col() calls in the script.

Step 3 — Generate the HTML

Use Python's string .replace() injection — not f-strings — because the template contains JavaScript ternary operators (?) that Python would try to evaluate as f-string expressions.

SKILL_DIR = "{baseDir}"

with open(f'{SKILL_DIR}/references/html_template.html') as f:
    html = f.read()
with open('node_modules/leaflet/dist/leaflet.js') as f:  leaflet_js  = f.read()
with open('node_modules/leaflet/dist/leaflet.css') as f: leaflet_css = f.read()
with open('outputs/imdf_b64.txt') as f:    imdf_b64 = f.read().strip()
with open('outputs/dash_data.json') as f:  va_data  = f.read()

# Customize these for the venue
VENUE_TITLE    = "One Bangkok Retail Data Navigator"
VENUE_SUBTITLE = "IMDF Indoor Map · Camera Groups · VA Analytics Overlay"
VENUE_EMOJI    = "🏢"
LS_KEY         = "oneBangkok_v1"   # localStorage key prefix — change per project

# Sensor color map: sensor_id -> hex color
# Generate automatically from VA data if not provided
import json
va = json.loads(va_data)
PALETTE = ['#1565C0','#2E7D32','#E65100','#C62828','#6A1B9A','#00838F',
           '#AD1457','#F57F17','#0277BD','#558B2F','#BF360C','#4527A0']
sensor_colors = {cid: PALETTE[i % len(PALETTE)] for i, cid in enumerate(va['cameras'])}

# Age group order (must match what's in the data)
age_order = ['Under 18','18-25','26-35','36-45','46-55','56-65','65+']

html = html.replace('__LEAFLET_CSS__',    leaflet_css)
html = html.replace('__LEAFLET_JS__',     leaflet_js)
html = html.replace('__IMDF_B64__',       imdf_b64)
html = html.replace('__VA_DATA__',        va_data)
html = html.replace('__VENUE_TITLE__',    VENUE_TITLE)
html = html.replace('__VENUE_SUBTITLE__', VENUE_SUBTITLE)
html = html.replace('"__VENUE_EMOJI__"',  f'"{VENUE_EMOJI}"')   # in JS
html = html.replace('__VENUE_EMOJI__',    VENUE_EMOJI)           # in HTML
html = html.replace('__LS_KEY__',         LS_KEY)
html = html.replace('__SENSOR_COLORS__',  json.dumps(sensor_colors))
html = html.replace('__AGE_ORDER__',      json.dumps(age_order))

with open('venue_map.html', 'w') as f:
    f.write(html)
print(f"Done — {len(html)//1024}KB")

Save the output HTML to the workspace folder so the user can open it.

Key technical patterns (know these to avoid bugs)

Per-floor marker visibility

Markers are NOT added to the map object directly on creation. They are added/removed in syncMarkerVisibility() when the floor changes. The rule is: if p.levelId === currentLevel → addTo(map). L.layerGroup() does not have .bringToBack() — do not call it.

Straight lines on GeoJSON

Set smoothFactor: 0 on every L.geoJSON() call. The template already does this inside CAT_STYLE and buildLevelLayer().

Unit category colors (CAT_STYLE)

The template uses a light/white color scheme with distinctive fill colors. Categories: room, walkway, restroom, escalator, elevator, stairs, nonpublic, opentobelow, default. Override in CAT_STYLE if the venue uses different category names.

Group data model

Groups use sensorIds (not cameraIds) as the array key in the JSON. Old v1 files used cameraIds — if importing old data add a migration shim.

localStorage keys

Use a unique LS_KEY per project so different venues don't share state. Format: <LS_KEY>_placements and <LS_KEY>_groups.

VA data structure expected by template

{
  "cameras": ["113034", "113042"],
  "cam_data": {
    "113034": {
      "total": 12500,
      "gender": {"MALE": 7000, "FEMALE": 5500},
      "age": {"Under 18": 200, "18-25": 3000, ...},
      "footfall_hour": {"0": 10, "1": 5, ..., "23": 400},
      "footfall_date": {"2026-06-17": 1800, ...},
      "footfall_dh": {"2026-06-17": {"8": 200, "9": 350, ...}}
    }
  }
}

Customisation checklist

• VENUE_TITLE — name shown in header and browser tab
• VENUE_SUBTITLE — subtitle line in header
• VENUE_EMOJI — emoji logo (e.g. 🏢 🛫 🛒 🏥)
• LS_KEY — unique localStorage prefix for this deployment
• sensor_colors — per-sensor colour palette (auto-generated if omitted)
• age_order — match the age labels in the actual data
• CAT_STYLE — adjust colours if venue uses non-standard unit categories
• EXCLUDE_UNIT_CATEGORIES in process_imdf.py — filter unwanted unit types

Troubleshooting

File size expectations

For very large venues (>1000 units), consider more aggressive unit filtering in process_imdf.py (e.g. filter nonpublic units too).